01 · Choose the implementation trade-off
Two separate accounts. Seven allowed tax budgets. The same illustrative prices, risk model, target and 5 bp transaction-cost assumption.
The tax-reserve / risk frontier
Each point is a solved budget scenario. Overlapping points mean a larger budget does not justify spending more under this fixed objective. Risk is a model estimate, not realized performance.
Exposure before and after
After weights use post-fee, post-reserve investable NAV. The optimizer's convex tracking objective uses starting-NAV-normalized dollar positions; the two definitions are deliberately shown separately.
All seven budget decisions and exact units
On small screens, scroll the table horizontally.
Proposed lot-level trades
Fractional shares; modeled basis excludes transaction fees. The policy sells initially overweight or prohibited names and buys initially underweight names. Losses receive no tax credit. No wash-sale or whole-share execution model is implied.
02 · Challenge the instruction
A separate implementation reconstructs sales, purchases, fees, positive-gain reserves, ending cash and positions. It does not call the optimizer's accounting helpers.
These are saved outcomes from executable fault tests, not controls running on a live portfolio. Payload fingerprints were refreshed for economic fault tests so a hash check could not hide a missing accounting control.
03 · Recover without applying a trade twice
The paper journal uses atomic SQLite transactions, explicit state versions and idempotency keys. A contribution changes capital, not investment profit.
One hypothetical decision date. A reopen-and-replay demonstration is evidence of tested recovery behavior, not a claim of production uptime or long-running monitoring.
04 · What the evidence supports
All account values, tax rates, prices and covariance inputs are invented for this lab. The separate historical WRDS/CRSP study remains unchanged. A technical PASS releases a paper instruction for review; it does not authorize a trade.